A thermal deformation estimation method for high precision machine tool spindles based on the convolutional neural network

The temperature rise of key internal components and ambient temperature changes during machining processes are the main causes of thermal displacement in machine tool spindle. This wastes materials and reduces working efficiency. A highly accurate and robust thermal spindle displacement estimation scheme is proposed in this paper which is based on the convolutional neural network (CNN) technique. Several key signals of specific dimensions related to spindle thermal displacement were first collected as two-dimensional (2D) signal maps. A multi-level feature expression for these 2D signal maps was extracted using convolution and pooling. A relationship between the extracted features and spindle thermal displacement was then learned using the neural architecture of the full connection layer. Optimized hyperparameter settings were determined by design of experiments (DoE) applied to the proposed CNN model. Experimental results showed that the proposed method had better performance than the multiple regression analysis (MLR) or back propagation neural network (BPNN) methods in terms of estimation accuracy and robustness at different spindle speeds.